[Mesa-dev] [PATCH 03/10] intel/fs: Fix bug in lower_simd_width while splitting an instruction which was already split.

Francisco Jerez currojerez at riseup.net
Sat Dec 29 20:38:57 UTC 2018


This seems to be a problem in combination with the lower_regioning
pass introduced by a future commit, which can modify a SIMD-split
instruction causing its execution size to become illegal again.  A
subsequent call to lower_simd_width() would hit this bug on a future
platform.

Cc: mesa-stable at lists.freedesktop.org
---
 src/intel/compiler/brw_fs.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index 97544fdf465..4aacc72a1b7 100644
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -5666,7 +5666,7 @@ static fs_reg
 emit_unzip(const fs_builder &lbld, fs_inst *inst, unsigned i)
 {
    /* Specified channel group from the source region. */
-   const fs_reg src = horiz_offset(inst->src[i], lbld.group());
+   const fs_reg src = horiz_offset(inst->src[i], lbld.group() - inst->group);
 
    if (needs_src_copy(lbld, inst, i)) {
       /* Builder of the right width to perform the copy avoiding uninitialized
@@ -5757,7 +5757,7 @@ emit_zip(const fs_builder &lbld_before, const fs_builder &lbld_after,
    assert(lbld_before.group() == lbld_after.group());
 
    /* Specified channel group from the destination region. */
-   const fs_reg dst = horiz_offset(inst->dst, lbld_after.group());
+   const fs_reg dst = horiz_offset(inst->dst, lbld_after.group() - inst->group);
    const unsigned dst_size = inst->size_written /
       inst->dst.component_size(inst->exec_size);
 
-- 
2.19.2



More information about the mesa-dev mailing list